AutoDDG
收藏资源简介:
AutoDDG是一个针对表格数据的自动数据集描述生成框架,旨在提高基于关键词的数据集搜索效果。框架结合了数据驱动方法和大型语言模型,生成全面、准确、易读且简洁的数据集描述。该框架首先通过传统数据配置文件技术生成数据驱动配置文件,捕获数据集的基本特征,然后利用大型语言模型衍生语义配置文件,丰富描述的上下文信息。AutoDDG生成的描述在数据检索、完整性和一致性方面进行了评估,并通过两个新的基准测试进行验证,这些基准测试专门设计用于评估数据集描述的质量。
AutoDDG is an automated dataset description generation framework tailored for tabular data, aiming to improve keyword-based dataset search performance. It combines data-driven methods and Large Language Models (LLMs) to generate comprehensive, accurate, readable and concise dataset descriptions. This framework first generates data-driven configuration files via traditional data profile techniques to capture the fundamental characteristics of datasets, then leverages LLMs to derive semantic configuration files that enrich the contextual information of the descriptions. The descriptions generated by AutoDDG are evaluated in terms of data retrieval, completeness and consistency, and validated through two novel benchmarks specifically designed to assess the quality of dataset descriptions.




